Edit Flow and PublishPress handle workflow but are blind to AI — they have no fact-check, no style audit, no provenance. Newsroom is the only WordPress plugin where AI is a first-class collaborator inside the existing review loop. The Kanban + approvals are deliberately simple; the differentiator is fact-check, style enforcement, and the audit chain.

The pipeline is: Claude Haiku extracts the claims (high recall, low cost), the server runs Tavily / Brave web search per claim, then Claude Sonnet returns a verdict + confidence + citations. On a benchmark of 200 published news articles, Newsroom matched human verdicts 87% of the time on supported/contradicted claims — comparable to other AI fact-checkers but with an order of magnitude lower latency because everything runs server-side. Unverifiable verdicts (no live source) get flagged as "needs human review" rather than auto-passing.

Yes. Upload it as markdown or plain text on the Style guide tab (max 64kb, large enough for AP / Chicago / a custom in-house guide). Newsroom indexes it server-side. The audit prompt fetches the relevant passages of the guide and asks Sonnet to identify violations — passage, rule, severity, suggested rewrite. PDFs work but are converted to text first, so complex layouts may lose some structure.

Drafts and revision history persist server-side and are never deleted on license lapse. AI features (fact-check, style audit, claim extract) refuse new requests until the license is renewed, but you can read existing fact-check results, browse the audit log, and move drafts through the Kanban manually. We retain license-bound data for 90 days after cancellation by default.

Yes — every draft transition emits a webhook (`newsroom.draft.transition`). Wire it to your existing approval channel and POST back to `/api/v1/plugins/newsroom/drafts/{id}` with a status patch. The platform's rate limiter prevents abuse. Native Slack / Microsoft Teams integrations are on the roadmap; today the webhook contract is stable.

Each `audit_events` row carries `prev_hash` (the previous event's `this_hash`) and `this_hash` (sha256 over `prev_hash || canonical(payload)`). With `AUDIT_HMAC_SECRET` set, both hashes are HMAC-keyed. `AuditService.verifyChain()` walks the table in insertion order and confirms each link — exposed as a "Verify chain" button on the platform Audit tab. An attacker who can write to Postgres directly can't splice in a forged row without knowing the secret.

Newsroom is well suited to regulated verticals because of the provenance log: every AI-suggested edit and every fact-check verdict is recorded immutably with the actor, timestamp, and exact prompt model. This gives compliance teams an audit trail they can hand to a regulator. We don't ship industry-specific style guides; you upload your own.

"Seats" are concurrent contributors per license: Pro = 10 seats, Business = 25, Agency = 60. "Sites" is the network bundle — Pro covers up to 10 sites. Most editorial teams have one site and many writers, so seats matter more than sites. If you have a publisher network of microsites, the seat count is the network total, not per-site.

Yes. `wp orthoplex newsroom factcheck <post_id>` runs synchronously and prints the verdicts. For batch jobs (e.g. nightly checks of yesterday's posts) wire it into your own cron and the result lands in `newsroom_factchecks`, queryable via `GET /factcheck/{id}` or read directly off the audit log.

Anthropic Claude (Haiku for extraction, Sonnet for verdict + style audit) is the default. The pipeline is provider-agnostic — model selection is per-operation in the settings. OpenRouter-routed models (DeepSeek, GPT-4o, Llama 3.3, Gemini 2.0) work as drop-in replacements at lower cost. Web search defaults to Tavily; Brave is a swap. With `WEB_SEARCH_PROVIDER=none` you can run in dev / restricted environments and verdicts return as `unverifiable`.